Locally Adaptive Text Classification based k-nearest Neighbors

Due to the exponential growth of documents on the Internet and the emergent need to organize them, the automated categorization of documents into predefined labels has received an ever-increased attention in the recent years. Among all these classifiers, K-Nearest Neighbors (KNNC) is a widely used classifier in text categorization community because of its simplicity and efficiency. However, KNNC still suffers from inductive biases or model misfits that result from its assumptions, such as the presumption that training data are evenly distributed among all categories. In this paper, we propose a new refinement strategy (LAKNNC) for the KNN Classifier, which adopts sum-of-squared-error criterion to adaptively select the contributing part from these neighbors and classifies the input document in term of the disturbance degree which it brings to the kernel densities of these selected neighbors. The experimental results indicate that our algorithm LAKNNC is not sensitive to the parameter k and achieves significant classification performance improvement on imbalanced corpora.